Predictor variables that affect a building energy load are sampled at a plurality of times within a time period and aggregated to generate an aggregated value for each predictor variable over the time period. A model is generated which estimates the building energy load in terms of the predictor variables. A regression process is performed to generate values for a plurality of regression coefficients in the model based on a cumulative building energy load for the time period and the aggregated values. The sampled values of the predictor variables are then applied as inputs to the model to estimate building energy loads at each of the plurality of times. The estimated building energy loads may be used as inputs to a controller that optimizes operation of a central plant.